arXiv stat.ML
· Papers
Orthogonal Discrepancy Kernels for Learning with Partial Physics
arXiv:2606.21199v2 Announce Type: replace Abstract: We introduce a semi-parametric framework for nonlinear system identification, which decouples discrepancy functions from physics-based components. Orthogonal Gaussian process regression balances sparse parameter selection (the white box) with discrepancy learning (the